AI companies are tightening token limits. The last one to blink may win

For years, AI companies gave users unfettered access to the candy, encouraging them to think of tokens, the chunks of text AI reads and writes, as effectively infinite. Tokens were bundled into subscriptions, hidden behind generous caps, or priced low enough that people stopped counting them. But as the cost of serving models eats into revenue, and as chip shortages, helium disruption, and data center bottlenecks constrain how much compute can come online, the big model makers are starting to ration access aggressively. All-you-can-eat AI is disappearing.